1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * kernel/sched/debug.c 4 * 5 * Print the CFS rbtree and other debugging details 6 * 7 * Copyright(C) 2007, Red Hat, Inc., Ingo Molnar 8 */ 9 10 /* 11 * This allows printing both to /sys/kernel/debug/sched/debug and 12 * to the console 13 */ 14 #define SEQ_printf(m, x...) \ 15 do { \ 16 if (m) \ 17 seq_printf(m, x); \ 18 else \ 19 pr_cont(x); \ 20 } while (0) 21 22 /* 23 * Ease the printing of nsec fields: 24 */ 25 static long long nsec_high(unsigned long long nsec) 26 { 27 if ((long long)nsec < 0) { 28 nsec = -nsec; 29 do_div(nsec, 1000000); 30 return -nsec; 31 } 32 do_div(nsec, 1000000); 33 34 return nsec; 35 } 36 37 static unsigned long nsec_low(unsigned long long nsec) 38 { 39 if ((long long)nsec < 0) 40 nsec = -nsec; 41 42 return do_div(nsec, 1000000); 43 } 44 45 #define SPLIT_NS(x) nsec_high(x), nsec_low(x) 46 47 #define SCHED_FEAT(name, enabled) \ 48 #name , 49 50 static const char * const sched_feat_names[] = { 51 #include "features.h" 52 }; 53 54 #undef SCHED_FEAT 55 56 static int sched_feat_show(struct seq_file *m, void *v) 57 { 58 int i; 59 60 for (i = 0; i < __SCHED_FEAT_NR; i++) { 61 if (!(sysctl_sched_features & (1UL << i))) 62 seq_puts(m, "NO_"); 63 seq_printf(m, "%s ", sched_feat_names[i]); 64 } 65 seq_puts(m, "\n"); 66 67 return 0; 68 } 69 70 #ifdef CONFIG_JUMP_LABEL 71 72 #define jump_label_key__true STATIC_KEY_INIT_TRUE 73 #define jump_label_key__false STATIC_KEY_INIT_FALSE 74 75 #define SCHED_FEAT(name, enabled) \ 76 jump_label_key__##enabled , 77 78 struct static_key sched_feat_keys[__SCHED_FEAT_NR] = { 79 #include "features.h" 80 }; 81 82 #undef SCHED_FEAT 83 84 static void sched_feat_disable(int i) 85 { 86 static_key_disable_cpuslocked(&sched_feat_keys[i]); 87 } 88 89 static void sched_feat_enable(int i) 90 { 91 static_key_enable_cpuslocked(&sched_feat_keys[i]); 92 } 93 #else 94 static void sched_feat_disable(int i) { }; 95 static void sched_feat_enable(int i) { }; 96 #endif /* CONFIG_JUMP_LABEL */ 97 98 static int sched_feat_set(char *cmp) 99 { 100 int i; 101 int neg = 0; 102 103 if (strncmp(cmp, "NO_", 3) == 0) { 104 neg = 1; 105 cmp += 3; 106 } 107 108 i = match_string(sched_feat_names, __SCHED_FEAT_NR, cmp); 109 if (i < 0) 110 return i; 111 112 if (neg) { 113 sysctl_sched_features &= ~(1UL << i); 114 sched_feat_disable(i); 115 } else { 116 sysctl_sched_features |= (1UL << i); 117 sched_feat_enable(i); 118 } 119 120 return 0; 121 } 122 123 static ssize_t 124 sched_feat_write(struct file *filp, const char __user *ubuf, 125 size_t cnt, loff_t *ppos) 126 { 127 char buf[64]; 128 char *cmp; 129 int ret; 130 struct inode *inode; 131 132 if (cnt > 63) 133 cnt = 63; 134 135 if (copy_from_user(&buf, ubuf, cnt)) 136 return -EFAULT; 137 138 buf[cnt] = 0; 139 cmp = strstrip(buf); 140 141 /* Ensure the static_key remains in a consistent state */ 142 inode = file_inode(filp); 143 cpus_read_lock(); 144 inode_lock(inode); 145 ret = sched_feat_set(cmp); 146 inode_unlock(inode); 147 cpus_read_unlock(); 148 if (ret < 0) 149 return ret; 150 151 *ppos += cnt; 152 153 return cnt; 154 } 155 156 static int sched_feat_open(struct inode *inode, struct file *filp) 157 { 158 return single_open(filp, sched_feat_show, NULL); 159 } 160 161 static const struct file_operations sched_feat_fops = { 162 .open = sched_feat_open, 163 .write = sched_feat_write, 164 .read = seq_read, 165 .llseek = seq_lseek, 166 .release = single_release, 167 }; 168 169 #ifdef CONFIG_SMP 170 171 static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf, 172 size_t cnt, loff_t *ppos) 173 { 174 char buf[16]; 175 unsigned int scaling; 176 177 if (cnt > 15) 178 cnt = 15; 179 180 if (copy_from_user(&buf, ubuf, cnt)) 181 return -EFAULT; 182 buf[cnt] = '\0'; 183 184 if (kstrtouint(buf, 10, &scaling)) 185 return -EINVAL; 186 187 if (scaling >= SCHED_TUNABLESCALING_END) 188 return -EINVAL; 189 190 sysctl_sched_tunable_scaling = scaling; 191 if (sched_update_scaling()) 192 return -EINVAL; 193 194 *ppos += cnt; 195 return cnt; 196 } 197 198 static int sched_scaling_show(struct seq_file *m, void *v) 199 { 200 seq_printf(m, "%d\n", sysctl_sched_tunable_scaling); 201 return 0; 202 } 203 204 static int sched_scaling_open(struct inode *inode, struct file *filp) 205 { 206 return single_open(filp, sched_scaling_show, NULL); 207 } 208 209 static const struct file_operations sched_scaling_fops = { 210 .open = sched_scaling_open, 211 .write = sched_scaling_write, 212 .read = seq_read, 213 .llseek = seq_lseek, 214 .release = single_release, 215 }; 216 217 #endif /* SMP */ 218 219 #ifdef CONFIG_PREEMPT_DYNAMIC 220 221 static ssize_t sched_dynamic_write(struct file *filp, const char __user *ubuf, 222 size_t cnt, loff_t *ppos) 223 { 224 char buf[16]; 225 int mode; 226 227 if (cnt > 15) 228 cnt = 15; 229 230 if (copy_from_user(&buf, ubuf, cnt)) 231 return -EFAULT; 232 233 buf[cnt] = 0; 234 mode = sched_dynamic_mode(strstrip(buf)); 235 if (mode < 0) 236 return mode; 237 238 sched_dynamic_update(mode); 239 240 *ppos += cnt; 241 242 return cnt; 243 } 244 245 static int sched_dynamic_show(struct seq_file *m, void *v) 246 { 247 static const char * preempt_modes[] = { 248 "none", "voluntary", "full" 249 }; 250 int i; 251 252 for (i = 0; i < ARRAY_SIZE(preempt_modes); i++) { 253 if (preempt_dynamic_mode == i) 254 seq_puts(m, "("); 255 seq_puts(m, preempt_modes[i]); 256 if (preempt_dynamic_mode == i) 257 seq_puts(m, ")"); 258 259 seq_puts(m, " "); 260 } 261 262 seq_puts(m, "\n"); 263 return 0; 264 } 265 266 static int sched_dynamic_open(struct inode *inode, struct file *filp) 267 { 268 return single_open(filp, sched_dynamic_show, NULL); 269 } 270 271 static const struct file_operations sched_dynamic_fops = { 272 .open = sched_dynamic_open, 273 .write = sched_dynamic_write, 274 .read = seq_read, 275 .llseek = seq_lseek, 276 .release = single_release, 277 }; 278 279 #endif /* CONFIG_PREEMPT_DYNAMIC */ 280 281 __read_mostly bool sched_debug_verbose; 282 283 #ifdef CONFIG_SMP 284 static struct dentry *sd_dentry; 285 286 287 static ssize_t sched_verbose_write(struct file *filp, const char __user *ubuf, 288 size_t cnt, loff_t *ppos) 289 { 290 ssize_t result; 291 bool orig; 292 293 cpus_read_lock(); 294 mutex_lock(&sched_domains_mutex); 295 296 orig = sched_debug_verbose; 297 result = debugfs_write_file_bool(filp, ubuf, cnt, ppos); 298 299 if (sched_debug_verbose && !orig) 300 update_sched_domain_debugfs(); 301 else if (!sched_debug_verbose && orig) { 302 debugfs_remove(sd_dentry); 303 sd_dentry = NULL; 304 } 305 306 mutex_unlock(&sched_domains_mutex); 307 cpus_read_unlock(); 308 309 return result; 310 } 311 #else 312 #define sched_verbose_write debugfs_write_file_bool 313 #endif 314 315 static const struct file_operations sched_verbose_fops = { 316 .read = debugfs_read_file_bool, 317 .write = sched_verbose_write, 318 .open = simple_open, 319 .llseek = default_llseek, 320 }; 321 322 static const struct seq_operations sched_debug_sops; 323 324 static int sched_debug_open(struct inode *inode, struct file *filp) 325 { 326 return seq_open(filp, &sched_debug_sops); 327 } 328 329 static const struct file_operations sched_debug_fops = { 330 .open = sched_debug_open, 331 .read = seq_read, 332 .llseek = seq_lseek, 333 .release = seq_release, 334 }; 335 336 enum dl_param { 337 DL_RUNTIME = 0, 338 DL_PERIOD, 339 }; 340 341 static unsigned long fair_server_period_max = (1UL << 22) * NSEC_PER_USEC; /* ~4 seconds */ 342 static unsigned long fair_server_period_min = (100) * NSEC_PER_USEC; /* 100 us */ 343 344 static ssize_t sched_fair_server_write(struct file *filp, const char __user *ubuf, 345 size_t cnt, loff_t *ppos, enum dl_param param) 346 { 347 long cpu = (long) ((struct seq_file *) filp->private_data)->private; 348 struct rq *rq = cpu_rq(cpu); 349 u64 runtime, period; 350 size_t err; 351 int retval; 352 u64 value; 353 354 err = kstrtoull_from_user(ubuf, cnt, 10, &value); 355 if (err) 356 return err; 357 358 scoped_guard (rq_lock_irqsave, rq) { 359 runtime = rq->fair_server.dl_runtime; 360 period = rq->fair_server.dl_period; 361 362 switch (param) { 363 case DL_RUNTIME: 364 if (runtime == value) 365 break; 366 runtime = value; 367 break; 368 case DL_PERIOD: 369 if (value == period) 370 break; 371 period = value; 372 break; 373 } 374 375 if (runtime > period || 376 period > fair_server_period_max || 377 period < fair_server_period_min) { 378 return -EINVAL; 379 } 380 381 if (rq->cfs.h_nr_running) { 382 update_rq_clock(rq); 383 dl_server_stop(&rq->fair_server); 384 } 385 386 retval = dl_server_apply_params(&rq->fair_server, runtime, period, 0); 387 if (retval) 388 cnt = retval; 389 390 if (!runtime) 391 printk_deferred("Fair server disabled in CPU %d, system may crash due to starvation.\n", 392 cpu_of(rq)); 393 394 if (rq->cfs.h_nr_running) 395 dl_server_start(&rq->fair_server); 396 } 397 398 *ppos += cnt; 399 return cnt; 400 } 401 402 static size_t sched_fair_server_show(struct seq_file *m, void *v, enum dl_param param) 403 { 404 unsigned long cpu = (unsigned long) m->private; 405 struct rq *rq = cpu_rq(cpu); 406 u64 value; 407 408 switch (param) { 409 case DL_RUNTIME: 410 value = rq->fair_server.dl_runtime; 411 break; 412 case DL_PERIOD: 413 value = rq->fair_server.dl_period; 414 break; 415 } 416 417 seq_printf(m, "%llu\n", value); 418 return 0; 419 420 } 421 422 static ssize_t 423 sched_fair_server_runtime_write(struct file *filp, const char __user *ubuf, 424 size_t cnt, loff_t *ppos) 425 { 426 return sched_fair_server_write(filp, ubuf, cnt, ppos, DL_RUNTIME); 427 } 428 429 static int sched_fair_server_runtime_show(struct seq_file *m, void *v) 430 { 431 return sched_fair_server_show(m, v, DL_RUNTIME); 432 } 433 434 static int sched_fair_server_runtime_open(struct inode *inode, struct file *filp) 435 { 436 return single_open(filp, sched_fair_server_runtime_show, inode->i_private); 437 } 438 439 static const struct file_operations fair_server_runtime_fops = { 440 .open = sched_fair_server_runtime_open, 441 .write = sched_fair_server_runtime_write, 442 .read = seq_read, 443 .llseek = seq_lseek, 444 .release = single_release, 445 }; 446 447 static ssize_t 448 sched_fair_server_period_write(struct file *filp, const char __user *ubuf, 449 size_t cnt, loff_t *ppos) 450 { 451 return sched_fair_server_write(filp, ubuf, cnt, ppos, DL_PERIOD); 452 } 453 454 static int sched_fair_server_period_show(struct seq_file *m, void *v) 455 { 456 return sched_fair_server_show(m, v, DL_PERIOD); 457 } 458 459 static int sched_fair_server_period_open(struct inode *inode, struct file *filp) 460 { 461 return single_open(filp, sched_fair_server_period_show, inode->i_private); 462 } 463 464 static const struct file_operations fair_server_period_fops = { 465 .open = sched_fair_server_period_open, 466 .write = sched_fair_server_period_write, 467 .read = seq_read, 468 .llseek = seq_lseek, 469 .release = single_release, 470 }; 471 472 static struct dentry *debugfs_sched; 473 474 static void debugfs_fair_server_init(void) 475 { 476 struct dentry *d_fair; 477 unsigned long cpu; 478 479 d_fair = debugfs_create_dir("fair_server", debugfs_sched); 480 if (!d_fair) 481 return; 482 483 for_each_possible_cpu(cpu) { 484 struct dentry *d_cpu; 485 char buf[32]; 486 487 snprintf(buf, sizeof(buf), "cpu%lu", cpu); 488 d_cpu = debugfs_create_dir(buf, d_fair); 489 490 debugfs_create_file("runtime", 0644, d_cpu, (void *) cpu, &fair_server_runtime_fops); 491 debugfs_create_file("period", 0644, d_cpu, (void *) cpu, &fair_server_period_fops); 492 } 493 } 494 495 static __init int sched_init_debug(void) 496 { 497 struct dentry __maybe_unused *numa; 498 499 debugfs_sched = debugfs_create_dir("sched", NULL); 500 501 debugfs_create_file("features", 0644, debugfs_sched, NULL, &sched_feat_fops); 502 debugfs_create_file_unsafe("verbose", 0644, debugfs_sched, &sched_debug_verbose, &sched_verbose_fops); 503 #ifdef CONFIG_PREEMPT_DYNAMIC 504 debugfs_create_file("preempt", 0644, debugfs_sched, NULL, &sched_dynamic_fops); 505 #endif 506 507 debugfs_create_u32("base_slice_ns", 0644, debugfs_sched, &sysctl_sched_base_slice); 508 509 debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms); 510 debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once); 511 512 #ifdef CONFIG_SMP 513 debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops); 514 debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost); 515 debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate); 516 517 mutex_lock(&sched_domains_mutex); 518 update_sched_domain_debugfs(); 519 mutex_unlock(&sched_domains_mutex); 520 #endif 521 522 #ifdef CONFIG_NUMA_BALANCING 523 numa = debugfs_create_dir("numa_balancing", debugfs_sched); 524 525 debugfs_create_u32("scan_delay_ms", 0644, numa, &sysctl_numa_balancing_scan_delay); 526 debugfs_create_u32("scan_period_min_ms", 0644, numa, &sysctl_numa_balancing_scan_period_min); 527 debugfs_create_u32("scan_period_max_ms", 0644, numa, &sysctl_numa_balancing_scan_period_max); 528 debugfs_create_u32("scan_size_mb", 0644, numa, &sysctl_numa_balancing_scan_size); 529 debugfs_create_u32("hot_threshold_ms", 0644, numa, &sysctl_numa_balancing_hot_threshold); 530 #endif 531 532 debugfs_create_file("debug", 0444, debugfs_sched, NULL, &sched_debug_fops); 533 534 debugfs_fair_server_init(); 535 536 return 0; 537 } 538 late_initcall(sched_init_debug); 539 540 #ifdef CONFIG_SMP 541 542 static cpumask_var_t sd_sysctl_cpus; 543 544 static int sd_flags_show(struct seq_file *m, void *v) 545 { 546 unsigned long flags = *(unsigned int *)m->private; 547 int idx; 548 549 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) { 550 seq_puts(m, sd_flag_debug[idx].name); 551 seq_puts(m, " "); 552 } 553 seq_puts(m, "\n"); 554 555 return 0; 556 } 557 558 static int sd_flags_open(struct inode *inode, struct file *file) 559 { 560 return single_open(file, sd_flags_show, inode->i_private); 561 } 562 563 static const struct file_operations sd_flags_fops = { 564 .open = sd_flags_open, 565 .read = seq_read, 566 .llseek = seq_lseek, 567 .release = single_release, 568 }; 569 570 static void register_sd(struct sched_domain *sd, struct dentry *parent) 571 { 572 #define SDM(type, mode, member) \ 573 debugfs_create_##type(#member, mode, parent, &sd->member) 574 575 SDM(ulong, 0644, min_interval); 576 SDM(ulong, 0644, max_interval); 577 SDM(u64, 0644, max_newidle_lb_cost); 578 SDM(u32, 0644, busy_factor); 579 SDM(u32, 0644, imbalance_pct); 580 SDM(u32, 0644, cache_nice_tries); 581 SDM(str, 0444, name); 582 583 #undef SDM 584 585 debugfs_create_file("flags", 0444, parent, &sd->flags, &sd_flags_fops); 586 debugfs_create_file("groups_flags", 0444, parent, &sd->groups->flags, &sd_flags_fops); 587 debugfs_create_u32("level", 0444, parent, (u32 *)&sd->level); 588 } 589 590 void update_sched_domain_debugfs(void) 591 { 592 int cpu, i; 593 594 /* 595 * This can unfortunately be invoked before sched_debug_init() creates 596 * the debug directory. Don't touch sd_sysctl_cpus until then. 597 */ 598 if (!debugfs_sched) 599 return; 600 601 if (!sched_debug_verbose) 602 return; 603 604 if (!cpumask_available(sd_sysctl_cpus)) { 605 if (!alloc_cpumask_var(&sd_sysctl_cpus, GFP_KERNEL)) 606 return; 607 cpumask_copy(sd_sysctl_cpus, cpu_possible_mask); 608 } 609 610 if (!sd_dentry) { 611 sd_dentry = debugfs_create_dir("domains", debugfs_sched); 612 613 /* rebuild sd_sysctl_cpus if empty since it gets cleared below */ 614 if (cpumask_empty(sd_sysctl_cpus)) 615 cpumask_copy(sd_sysctl_cpus, cpu_online_mask); 616 } 617 618 for_each_cpu(cpu, sd_sysctl_cpus) { 619 struct sched_domain *sd; 620 struct dentry *d_cpu; 621 char buf[32]; 622 623 snprintf(buf, sizeof(buf), "cpu%d", cpu); 624 debugfs_lookup_and_remove(buf, sd_dentry); 625 d_cpu = debugfs_create_dir(buf, sd_dentry); 626 627 i = 0; 628 for_each_domain(cpu, sd) { 629 struct dentry *d_sd; 630 631 snprintf(buf, sizeof(buf), "domain%d", i); 632 d_sd = debugfs_create_dir(buf, d_cpu); 633 634 register_sd(sd, d_sd); 635 i++; 636 } 637 638 __cpumask_clear_cpu(cpu, sd_sysctl_cpus); 639 } 640 } 641 642 void dirty_sched_domain_sysctl(int cpu) 643 { 644 if (cpumask_available(sd_sysctl_cpus)) 645 __cpumask_set_cpu(cpu, sd_sysctl_cpus); 646 } 647 648 #endif /* CONFIG_SMP */ 649 650 #ifdef CONFIG_FAIR_GROUP_SCHED 651 static void print_cfs_group_stats(struct seq_file *m, int cpu, struct task_group *tg) 652 { 653 struct sched_entity *se = tg->se[cpu]; 654 655 #define P(F) SEQ_printf(m, " .%-30s: %lld\n", #F, (long long)F) 656 #define P_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld\n", \ 657 #F, (long long)schedstat_val(stats->F)) 658 #define PN(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", #F, SPLIT_NS((long long)F)) 659 #define PN_SCHEDSTAT(F) SEQ_printf(m, " .%-30s: %lld.%06ld\n", \ 660 #F, SPLIT_NS((long long)schedstat_val(stats->F))) 661 662 if (!se) 663 return; 664 665 PN(se->exec_start); 666 PN(se->vruntime); 667 PN(se->sum_exec_runtime); 668 669 if (schedstat_enabled()) { 670 struct sched_statistics *stats; 671 stats = __schedstats_from_se(se); 672 673 PN_SCHEDSTAT(wait_start); 674 PN_SCHEDSTAT(sleep_start); 675 PN_SCHEDSTAT(block_start); 676 PN_SCHEDSTAT(sleep_max); 677 PN_SCHEDSTAT(block_max); 678 PN_SCHEDSTAT(exec_max); 679 PN_SCHEDSTAT(slice_max); 680 PN_SCHEDSTAT(wait_max); 681 PN_SCHEDSTAT(wait_sum); 682 P_SCHEDSTAT(wait_count); 683 } 684 685 P(se->load.weight); 686 #ifdef CONFIG_SMP 687 P(se->avg.load_avg); 688 P(se->avg.util_avg); 689 P(se->avg.runnable_avg); 690 #endif 691 692 #undef PN_SCHEDSTAT 693 #undef PN 694 #undef P_SCHEDSTAT 695 #undef P 696 } 697 #endif 698 699 #ifdef CONFIG_CGROUP_SCHED 700 static DEFINE_SPINLOCK(sched_debug_lock); 701 static char group_path[PATH_MAX]; 702 703 static void task_group_path(struct task_group *tg, char *path, int plen) 704 { 705 if (autogroup_path(tg, path, plen)) 706 return; 707 708 cgroup_path(tg->css.cgroup, path, plen); 709 } 710 711 /* 712 * Only 1 SEQ_printf_task_group_path() caller can use the full length 713 * group_path[] for cgroup path. Other simultaneous callers will have 714 * to use a shorter stack buffer. A "..." suffix is appended at the end 715 * of the stack buffer so that it will show up in case the output length 716 * matches the given buffer size to indicate possible path name truncation. 717 */ 718 #define SEQ_printf_task_group_path(m, tg, fmt...) \ 719 { \ 720 if (spin_trylock(&sched_debug_lock)) { \ 721 task_group_path(tg, group_path, sizeof(group_path)); \ 722 SEQ_printf(m, fmt, group_path); \ 723 spin_unlock(&sched_debug_lock); \ 724 } else { \ 725 char buf[128]; \ 726 char *bufend = buf + sizeof(buf) - 3; \ 727 task_group_path(tg, buf, bufend - buf); \ 728 strcpy(bufend - 1, "..."); \ 729 SEQ_printf(m, fmt, buf); \ 730 } \ 731 } 732 #endif 733 734 static void 735 print_task(struct seq_file *m, struct rq *rq, struct task_struct *p) 736 { 737 if (task_current(rq, p)) 738 SEQ_printf(m, ">R"); 739 else 740 SEQ_printf(m, " %c", task_state_to_char(p)); 741 742 SEQ_printf(m, " %15s %5d %9Ld.%06ld %c %9Ld.%06ld %c %9Ld.%06ld %9Ld.%06ld %9Ld %5d ", 743 p->comm, task_pid_nr(p), 744 SPLIT_NS(p->se.vruntime), 745 entity_eligible(cfs_rq_of(&p->se), &p->se) ? 'E' : 'N', 746 SPLIT_NS(p->se.deadline), 747 p->se.custom_slice ? 'S' : ' ', 748 SPLIT_NS(p->se.slice), 749 SPLIT_NS(p->se.sum_exec_runtime), 750 (long long)(p->nvcsw + p->nivcsw), 751 p->prio); 752 753 SEQ_printf(m, "%9lld.%06ld %9lld.%06ld %9lld.%06ld", 754 SPLIT_NS(schedstat_val_or_zero(p->stats.wait_sum)), 755 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_sleep_runtime)), 756 SPLIT_NS(schedstat_val_or_zero(p->stats.sum_block_runtime))); 757 758 #ifdef CONFIG_NUMA_BALANCING 759 SEQ_printf(m, " %d %d", task_node(p), task_numa_group_id(p)); 760 #endif 761 #ifdef CONFIG_CGROUP_SCHED 762 SEQ_printf_task_group_path(m, task_group(p), " %s") 763 #endif 764 765 SEQ_printf(m, "\n"); 766 } 767 768 static void print_rq(struct seq_file *m, struct rq *rq, int rq_cpu) 769 { 770 struct task_struct *g, *p; 771 772 SEQ_printf(m, "\n"); 773 SEQ_printf(m, "runnable tasks:\n"); 774 SEQ_printf(m, " S task PID vruntime eligible " 775 "deadline slice sum-exec switches " 776 "prio wait-time sum-sleep sum-block" 777 #ifdef CONFIG_NUMA_BALANCING 778 " node group-id" 779 #endif 780 #ifdef CONFIG_CGROUP_SCHED 781 " group-path" 782 #endif 783 "\n"); 784 SEQ_printf(m, "-------------------------------------------------------" 785 "------------------------------------------------------" 786 "------------------------------------------------------" 787 #ifdef CONFIG_NUMA_BALANCING 788 "--------------" 789 #endif 790 #ifdef CONFIG_CGROUP_SCHED 791 "--------------" 792 #endif 793 "\n"); 794 795 rcu_read_lock(); 796 for_each_process_thread(g, p) { 797 if (task_cpu(p) != rq_cpu) 798 continue; 799 800 print_task(m, rq, p); 801 } 802 rcu_read_unlock(); 803 } 804 805 void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq) 806 { 807 s64 left_vruntime = -1, min_vruntime, right_vruntime = -1, left_deadline = -1, spread; 808 struct sched_entity *last, *first, *root; 809 struct rq *rq = cpu_rq(cpu); 810 unsigned long flags; 811 812 #ifdef CONFIG_FAIR_GROUP_SCHED 813 SEQ_printf(m, "\n"); 814 SEQ_printf_task_group_path(m, cfs_rq->tg, "cfs_rq[%d]:%s\n", cpu); 815 #else 816 SEQ_printf(m, "\n"); 817 SEQ_printf(m, "cfs_rq[%d]:\n", cpu); 818 #endif 819 820 raw_spin_rq_lock_irqsave(rq, flags); 821 root = __pick_root_entity(cfs_rq); 822 if (root) 823 left_vruntime = root->min_vruntime; 824 first = __pick_first_entity(cfs_rq); 825 if (first) 826 left_deadline = first->deadline; 827 last = __pick_last_entity(cfs_rq); 828 if (last) 829 right_vruntime = last->vruntime; 830 min_vruntime = cfs_rq->min_vruntime; 831 raw_spin_rq_unlock_irqrestore(rq, flags); 832 833 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_deadline", 834 SPLIT_NS(left_deadline)); 835 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "left_vruntime", 836 SPLIT_NS(left_vruntime)); 837 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "min_vruntime", 838 SPLIT_NS(min_vruntime)); 839 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "avg_vruntime", 840 SPLIT_NS(avg_vruntime(cfs_rq))); 841 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "right_vruntime", 842 SPLIT_NS(right_vruntime)); 843 spread = right_vruntime - left_vruntime; 844 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", "spread", SPLIT_NS(spread)); 845 SEQ_printf(m, " .%-30s: %d\n", "nr_running", cfs_rq->nr_running); 846 SEQ_printf(m, " .%-30s: %d\n", "h_nr_running", cfs_rq->h_nr_running); 847 SEQ_printf(m, " .%-30s: %d\n", "idle_nr_running", 848 cfs_rq->idle_nr_running); 849 SEQ_printf(m, " .%-30s: %d\n", "idle_h_nr_running", 850 cfs_rq->idle_h_nr_running); 851 SEQ_printf(m, " .%-30s: %ld\n", "load", cfs_rq->load.weight); 852 #ifdef CONFIG_SMP 853 SEQ_printf(m, " .%-30s: %lu\n", "load_avg", 854 cfs_rq->avg.load_avg); 855 SEQ_printf(m, " .%-30s: %lu\n", "runnable_avg", 856 cfs_rq->avg.runnable_avg); 857 SEQ_printf(m, " .%-30s: %lu\n", "util_avg", 858 cfs_rq->avg.util_avg); 859 SEQ_printf(m, " .%-30s: %u\n", "util_est", 860 cfs_rq->avg.util_est); 861 SEQ_printf(m, " .%-30s: %ld\n", "removed.load_avg", 862 cfs_rq->removed.load_avg); 863 SEQ_printf(m, " .%-30s: %ld\n", "removed.util_avg", 864 cfs_rq->removed.util_avg); 865 SEQ_printf(m, " .%-30s: %ld\n", "removed.runnable_avg", 866 cfs_rq->removed.runnable_avg); 867 #ifdef CONFIG_FAIR_GROUP_SCHED 868 SEQ_printf(m, " .%-30s: %lu\n", "tg_load_avg_contrib", 869 cfs_rq->tg_load_avg_contrib); 870 SEQ_printf(m, " .%-30s: %ld\n", "tg_load_avg", 871 atomic_long_read(&cfs_rq->tg->load_avg)); 872 #endif 873 #endif 874 #ifdef CONFIG_CFS_BANDWIDTH 875 SEQ_printf(m, " .%-30s: %d\n", "throttled", 876 cfs_rq->throttled); 877 SEQ_printf(m, " .%-30s: %d\n", "throttle_count", 878 cfs_rq->throttle_count); 879 #endif 880 881 #ifdef CONFIG_FAIR_GROUP_SCHED 882 print_cfs_group_stats(m, cpu, cfs_rq->tg); 883 #endif 884 } 885 886 void print_rt_rq(struct seq_file *m, int cpu, struct rt_rq *rt_rq) 887 { 888 #ifdef CONFIG_RT_GROUP_SCHED 889 SEQ_printf(m, "\n"); 890 SEQ_printf_task_group_path(m, rt_rq->tg, "rt_rq[%d]:%s\n", cpu); 891 #else 892 SEQ_printf(m, "\n"); 893 SEQ_printf(m, "rt_rq[%d]:\n", cpu); 894 #endif 895 896 #define P(x) \ 897 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rt_rq->x)) 898 #define PU(x) \ 899 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(rt_rq->x)) 900 #define PN(x) \ 901 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rt_rq->x)) 902 903 PU(rt_nr_running); 904 905 #ifdef CONFIG_RT_GROUP_SCHED 906 P(rt_throttled); 907 PN(rt_time); 908 PN(rt_runtime); 909 #endif 910 911 #undef PN 912 #undef PU 913 #undef P 914 } 915 916 void print_dl_rq(struct seq_file *m, int cpu, struct dl_rq *dl_rq) 917 { 918 struct dl_bw *dl_bw; 919 920 SEQ_printf(m, "\n"); 921 SEQ_printf(m, "dl_rq[%d]:\n", cpu); 922 923 #define PU(x) \ 924 SEQ_printf(m, " .%-30s: %lu\n", #x, (unsigned long)(dl_rq->x)) 925 926 PU(dl_nr_running); 927 #ifdef CONFIG_SMP 928 dl_bw = &cpu_rq(cpu)->rd->dl_bw; 929 #else 930 dl_bw = &dl_rq->dl_bw; 931 #endif 932 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->bw", dl_bw->bw); 933 SEQ_printf(m, " .%-30s: %lld\n", "dl_bw->total_bw", dl_bw->total_bw); 934 935 #undef PU 936 } 937 938 static void print_cpu(struct seq_file *m, int cpu) 939 { 940 struct rq *rq = cpu_rq(cpu); 941 942 #ifdef CONFIG_X86 943 { 944 unsigned int freq = cpu_khz ? : 1; 945 946 SEQ_printf(m, "cpu#%d, %u.%03u MHz\n", 947 cpu, freq / 1000, (freq % 1000)); 948 } 949 #else 950 SEQ_printf(m, "cpu#%d\n", cpu); 951 #endif 952 953 #define P(x) \ 954 do { \ 955 if (sizeof(rq->x) == 4) \ 956 SEQ_printf(m, " .%-30s: %d\n", #x, (int)(rq->x)); \ 957 else \ 958 SEQ_printf(m, " .%-30s: %Ld\n", #x, (long long)(rq->x));\ 959 } while (0) 960 961 #define PN(x) \ 962 SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x)) 963 964 P(nr_running); 965 P(nr_switches); 966 P(nr_uninterruptible); 967 PN(next_balance); 968 SEQ_printf(m, " .%-30s: %ld\n", "curr->pid", (long)(task_pid_nr(rq->curr))); 969 PN(clock); 970 PN(clock_task); 971 #undef P 972 #undef PN 973 974 #ifdef CONFIG_SMP 975 #define P64(n) SEQ_printf(m, " .%-30s: %Ld\n", #n, rq->n); 976 P64(avg_idle); 977 P64(max_idle_balance_cost); 978 #undef P64 979 #endif 980 981 #define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, schedstat_val(rq->n)); 982 if (schedstat_enabled()) { 983 P(yld_count); 984 P(sched_count); 985 P(sched_goidle); 986 P(ttwu_count); 987 P(ttwu_local); 988 } 989 #undef P 990 991 print_cfs_stats(m, cpu); 992 print_rt_stats(m, cpu); 993 print_dl_stats(m, cpu); 994 995 print_rq(m, rq, cpu); 996 SEQ_printf(m, "\n"); 997 } 998 999 static const char *sched_tunable_scaling_names[] = { 1000 "none", 1001 "logarithmic", 1002 "linear" 1003 }; 1004 1005 static void sched_debug_header(struct seq_file *m) 1006 { 1007 u64 ktime, sched_clk, cpu_clk; 1008 unsigned long flags; 1009 1010 local_irq_save(flags); 1011 ktime = ktime_to_ns(ktime_get()); 1012 sched_clk = sched_clock(); 1013 cpu_clk = local_clock(); 1014 local_irq_restore(flags); 1015 1016 SEQ_printf(m, "Sched Debug Version: v0.11, %s %.*s\n", 1017 init_utsname()->release, 1018 (int)strcspn(init_utsname()->version, " "), 1019 init_utsname()->version); 1020 1021 #define P(x) \ 1022 SEQ_printf(m, "%-40s: %Ld\n", #x, (long long)(x)) 1023 #define PN(x) \ 1024 SEQ_printf(m, "%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) 1025 PN(ktime); 1026 PN(sched_clk); 1027 PN(cpu_clk); 1028 P(jiffies); 1029 #ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK 1030 P(sched_clock_stable()); 1031 #endif 1032 #undef PN 1033 #undef P 1034 1035 SEQ_printf(m, "\n"); 1036 SEQ_printf(m, "sysctl_sched\n"); 1037 1038 #define P(x) \ 1039 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x)) 1040 #define PN(x) \ 1041 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) 1042 PN(sysctl_sched_base_slice); 1043 P(sysctl_sched_features); 1044 #undef PN 1045 #undef P 1046 1047 SEQ_printf(m, " .%-40s: %d (%s)\n", 1048 "sysctl_sched_tunable_scaling", 1049 sysctl_sched_tunable_scaling, 1050 sched_tunable_scaling_names[sysctl_sched_tunable_scaling]); 1051 SEQ_printf(m, "\n"); 1052 } 1053 1054 static int sched_debug_show(struct seq_file *m, void *v) 1055 { 1056 int cpu = (unsigned long)(v - 2); 1057 1058 if (cpu != -1) 1059 print_cpu(m, cpu); 1060 else 1061 sched_debug_header(m); 1062 1063 return 0; 1064 } 1065 1066 void sysrq_sched_debug_show(void) 1067 { 1068 int cpu; 1069 1070 sched_debug_header(NULL); 1071 for_each_online_cpu(cpu) { 1072 /* 1073 * Need to reset softlockup watchdogs on all CPUs, because 1074 * another CPU might be blocked waiting for us to process 1075 * an IPI or stop_machine. 1076 */ 1077 touch_nmi_watchdog(); 1078 touch_all_softlockup_watchdogs(); 1079 print_cpu(NULL, cpu); 1080 } 1081 } 1082 1083 /* 1084 * This iterator needs some explanation. 1085 * It returns 1 for the header position. 1086 * This means 2 is CPU 0. 1087 * In a hotplugged system some CPUs, including CPU 0, may be missing so we have 1088 * to use cpumask_* to iterate over the CPUs. 1089 */ 1090 static void *sched_debug_start(struct seq_file *file, loff_t *offset) 1091 { 1092 unsigned long n = *offset; 1093 1094 if (n == 0) 1095 return (void *) 1; 1096 1097 n--; 1098 1099 if (n > 0) 1100 n = cpumask_next(n - 1, cpu_online_mask); 1101 else 1102 n = cpumask_first(cpu_online_mask); 1103 1104 *offset = n + 1; 1105 1106 if (n < nr_cpu_ids) 1107 return (void *)(unsigned long)(n + 2); 1108 1109 return NULL; 1110 } 1111 1112 static void *sched_debug_next(struct seq_file *file, void *data, loff_t *offset) 1113 { 1114 (*offset)++; 1115 return sched_debug_start(file, offset); 1116 } 1117 1118 static void sched_debug_stop(struct seq_file *file, void *data) 1119 { 1120 } 1121 1122 static const struct seq_operations sched_debug_sops = { 1123 .start = sched_debug_start, 1124 .next = sched_debug_next, 1125 .stop = sched_debug_stop, 1126 .show = sched_debug_show, 1127 }; 1128 1129 #define __PS(S, F) SEQ_printf(m, "%-45s:%21Ld\n", S, (long long)(F)) 1130 #define __P(F) __PS(#F, F) 1131 #define P(F) __PS(#F, p->F) 1132 #define PM(F, M) __PS(#F, p->F & (M)) 1133 #define __PSN(S, F) SEQ_printf(m, "%-45s:%14Ld.%06ld\n", S, SPLIT_NS((long long)(F))) 1134 #define __PN(F) __PSN(#F, F) 1135 #define PN(F) __PSN(#F, p->F) 1136 1137 1138 #ifdef CONFIG_NUMA_BALANCING 1139 void print_numa_stats(struct seq_file *m, int node, unsigned long tsf, 1140 unsigned long tpf, unsigned long gsf, unsigned long gpf) 1141 { 1142 SEQ_printf(m, "numa_faults node=%d ", node); 1143 SEQ_printf(m, "task_private=%lu task_shared=%lu ", tpf, tsf); 1144 SEQ_printf(m, "group_private=%lu group_shared=%lu\n", gpf, gsf); 1145 } 1146 #endif 1147 1148 1149 static void sched_show_numa(struct task_struct *p, struct seq_file *m) 1150 { 1151 #ifdef CONFIG_NUMA_BALANCING 1152 if (p->mm) 1153 P(mm->numa_scan_seq); 1154 1155 P(numa_pages_migrated); 1156 P(numa_preferred_nid); 1157 P(total_numa_faults); 1158 SEQ_printf(m, "current_node=%d, numa_group_id=%d\n", 1159 task_node(p), task_numa_group_id(p)); 1160 show_numa_stats(p, m); 1161 #endif 1162 } 1163 1164 void proc_sched_show_task(struct task_struct *p, struct pid_namespace *ns, 1165 struct seq_file *m) 1166 { 1167 unsigned long nr_switches; 1168 1169 SEQ_printf(m, "%s (%d, #threads: %d)\n", p->comm, task_pid_nr_ns(p, ns), 1170 get_nr_threads(p)); 1171 SEQ_printf(m, 1172 "---------------------------------------------------------" 1173 "----------\n"); 1174 1175 #define P_SCHEDSTAT(F) __PS(#F, schedstat_val(p->stats.F)) 1176 #define PN_SCHEDSTAT(F) __PSN(#F, schedstat_val(p->stats.F)) 1177 1178 PN(se.exec_start); 1179 PN(se.vruntime); 1180 PN(se.sum_exec_runtime); 1181 1182 nr_switches = p->nvcsw + p->nivcsw; 1183 1184 P(se.nr_migrations); 1185 1186 if (schedstat_enabled()) { 1187 u64 avg_atom, avg_per_cpu; 1188 1189 PN_SCHEDSTAT(sum_sleep_runtime); 1190 PN_SCHEDSTAT(sum_block_runtime); 1191 PN_SCHEDSTAT(wait_start); 1192 PN_SCHEDSTAT(sleep_start); 1193 PN_SCHEDSTAT(block_start); 1194 PN_SCHEDSTAT(sleep_max); 1195 PN_SCHEDSTAT(block_max); 1196 PN_SCHEDSTAT(exec_max); 1197 PN_SCHEDSTAT(slice_max); 1198 PN_SCHEDSTAT(wait_max); 1199 PN_SCHEDSTAT(wait_sum); 1200 P_SCHEDSTAT(wait_count); 1201 PN_SCHEDSTAT(iowait_sum); 1202 P_SCHEDSTAT(iowait_count); 1203 P_SCHEDSTAT(nr_migrations_cold); 1204 P_SCHEDSTAT(nr_failed_migrations_affine); 1205 P_SCHEDSTAT(nr_failed_migrations_running); 1206 P_SCHEDSTAT(nr_failed_migrations_hot); 1207 P_SCHEDSTAT(nr_forced_migrations); 1208 P_SCHEDSTAT(nr_wakeups); 1209 P_SCHEDSTAT(nr_wakeups_sync); 1210 P_SCHEDSTAT(nr_wakeups_migrate); 1211 P_SCHEDSTAT(nr_wakeups_local); 1212 P_SCHEDSTAT(nr_wakeups_remote); 1213 P_SCHEDSTAT(nr_wakeups_affine); 1214 P_SCHEDSTAT(nr_wakeups_affine_attempts); 1215 P_SCHEDSTAT(nr_wakeups_passive); 1216 P_SCHEDSTAT(nr_wakeups_idle); 1217 1218 avg_atom = p->se.sum_exec_runtime; 1219 if (nr_switches) 1220 avg_atom = div64_ul(avg_atom, nr_switches); 1221 else 1222 avg_atom = -1LL; 1223 1224 avg_per_cpu = p->se.sum_exec_runtime; 1225 if (p->se.nr_migrations) { 1226 avg_per_cpu = div64_u64(avg_per_cpu, 1227 p->se.nr_migrations); 1228 } else { 1229 avg_per_cpu = -1LL; 1230 } 1231 1232 __PN(avg_atom); 1233 __PN(avg_per_cpu); 1234 1235 #ifdef CONFIG_SCHED_CORE 1236 PN_SCHEDSTAT(core_forceidle_sum); 1237 #endif 1238 } 1239 1240 __P(nr_switches); 1241 __PS("nr_voluntary_switches", p->nvcsw); 1242 __PS("nr_involuntary_switches", p->nivcsw); 1243 1244 P(se.load.weight); 1245 #ifdef CONFIG_SMP 1246 P(se.avg.load_sum); 1247 P(se.avg.runnable_sum); 1248 P(se.avg.util_sum); 1249 P(se.avg.load_avg); 1250 P(se.avg.runnable_avg); 1251 P(se.avg.util_avg); 1252 P(se.avg.last_update_time); 1253 PM(se.avg.util_est, ~UTIL_AVG_UNCHANGED); 1254 #endif 1255 #ifdef CONFIG_UCLAMP_TASK 1256 __PS("uclamp.min", p->uclamp_req[UCLAMP_MIN].value); 1257 __PS("uclamp.max", p->uclamp_req[UCLAMP_MAX].value); 1258 __PS("effective uclamp.min", uclamp_eff_value(p, UCLAMP_MIN)); 1259 __PS("effective uclamp.max", uclamp_eff_value(p, UCLAMP_MAX)); 1260 #endif 1261 P(policy); 1262 P(prio); 1263 if (task_has_dl_policy(p)) { 1264 P(dl.runtime); 1265 P(dl.deadline); 1266 } 1267 #ifdef CONFIG_SCHED_CLASS_EXT 1268 __PS("ext.enabled", task_on_scx(p)); 1269 #endif 1270 #undef PN_SCHEDSTAT 1271 #undef P_SCHEDSTAT 1272 1273 { 1274 unsigned int this_cpu = raw_smp_processor_id(); 1275 u64 t0, t1; 1276 1277 t0 = cpu_clock(this_cpu); 1278 t1 = cpu_clock(this_cpu); 1279 __PS("clock-delta", t1-t0); 1280 } 1281 1282 sched_show_numa(p, m); 1283 } 1284 1285 void proc_sched_set_task(struct task_struct *p) 1286 { 1287 #ifdef CONFIG_SCHEDSTATS 1288 memset(&p->stats, 0, sizeof(p->stats)); 1289 #endif 1290 } 1291 1292 void resched_latency_warn(int cpu, u64 latency) 1293 { 1294 static DEFINE_RATELIMIT_STATE(latency_check_ratelimit, 60 * 60 * HZ, 1); 1295 1296 WARN(__ratelimit(&latency_check_ratelimit), 1297 "sched: CPU %d need_resched set for > %llu ns (%d ticks) " 1298 "without schedule\n", 1299 cpu, latency, cpu_rq(cpu)->ticks_without_resched); 1300 } 1301